The effect of type 2 diabetes diagnosis in the elderly

Economics and Human Biology 37 (2020) 100830

Contents lists available at ScienceDirect

Economics and Human Biology journal homepage: www.elsevier.com/locate/ehb

The effect of type 2 diabetes diagnosis in the elderly Alessio Gaggero, PhD University of Granada, Department of Applied Economics, Cartuja Campus, 18011, Granada, Spain

A R T I C L E I N F O

A B S T R A C T

Article history: Received 30 June 2019 Received in revised form 11 November 2019 Accepted 15 November 2019 Available online 21 February 2020

In this paper, I use a novel approach based on biomarkers data to examine the effect of type 2 diabetes (T2D) diagnosis on both physical and mental health for a sample of individuals aged 50 and above. In order to retrieve reliable estimates, I exploit the fact that medical guidelines cause a discontinuity in the probability that General Practitioners (GP) diagnose their patients with T2D as a function of exact blood test cut-off in a regression discontinuity design (RDD) approach. Using data from the English Longitudinal Study of Ageing (ELSA), I find compelling evidence that health information, in the form of T2D diagnosis, influences the protection of health into old age. Specifically, after receiving T2D diagnosis, over time, individuals reported a 2.1 lower body mass index (BMI) as well as 5.5 cm lower waist circumference, relative to their counterparts. With respect to self-reported physical health, the results imply that those diagnosed with T2D reported a 0.19 and 0.70 higher score in increased activities of daily living (ADL) and body mobility indexes, respectively. Finally, while I find evidence that T2D-diagnosed patients reported a significant, 0.5 higher, score in the word listening test, I find no evidence that T2D diagnosis impacted self-reported depression levels. I provide a wide variety of evidence on the validity of the results. © 2020 Elsevier B.V. All rights reserved.

Keywords: Biomarkers Type 2 diabetes diagnosis RDD

1. Introduction and motivation Everywhere around the world, life expectancy has increased dramatically. It is the first time in the history of human kind that most people can expect to live beyond the age of 60. The World Health Organisation (WHO, 2016) reports 900 million people are aged 60 and over, and this number is forecasted to become 2 billion by 2050.1 Since a natural component of the ageing process is the decline of both physical and mental health (Milanovi
c et al., 2013; Mazzonna and Peracchi, 2012), the rise in the elderly population is likely to create significant pressure on public finances, especially in terms of healthcare expenditure. Amongst the various chronic diseases affecting the ageing population, type 2 diabetes (T2D) is the most common. The British Geriatrics Society (2009) reports half of all people in the UK diagnosed with T2D are aged over 65 years, and also that, for them, the management of this condition is particularly challenging due to their frailty and tolerance to standard therapies.2 Despite this, there is still little evidence of the

1 2

E-mail address: [email protected] (A. Gaggero). https://www.who.int/features/factfiles/ageing/en/ https://www.bgs.org.uk/resources/diabetes

http://dx.doi.org/10.1016/j.ehb.2019.100830 1570-677X/© 2020 Elsevier B.V. All rights reserved.

effect that the diagnosis of T2D may have on this group of the population.3 Several studies show that individuals significantly change their lifestyle behaviours when diagnosed with T2D, in terms of diet, physical activity, smoking and drinking behaviours. For example, Slade (2012) explores the effect of T2D on lifestyle behaviours (weight, exercises, drinking and smoking) using a health and retirement survey with dynamic panel techniques. He finds that individuals initially responded to the T2D diagnosis by increasing exercise, losing weight, and curbing smoking and drinking behaviour, but that the effects diminish with time. Similar to this paper, Kim et al. (2019) use discontinuities for T2D, as well as for hyperlipidaemia (high cholesterol) and obesity diagnosis, to examine whether individuals change their lifestyle behaviours (weight and exercise) upon receiving the diagnosis. Employing a sample of the Korean population, they find evidence of weight loss around the threshold for T2D.4 5 Very little is known, however,

3 Several studies report the positive effects of following healthy lifestyles on a variety of health outcomes (Kenkel, 1995; Contoyannis and Jones, 2004; Balia and Jones, 2008). Further, there is substantial literature that directly links health to labour market outcomes (Cawley, 2015; Han et al., 2008; Lindeboom et al., 2010; Morris, 2007). 4 A number of papers report comparable changes in lifestyle behaviour following hypertension and high cholesterol diagnosis (Zhao et al., 2013; Gaggero, 2019). 5 There is some evidence of the negative effect of T2D diagnosis on employment (Seuring et al., 2015) and other labour market outcomes (Seuring et al., 2019).

2

A. Gaggero / Economics and Human Biology 37 (2020) 100830

concerning whether these changes in lifestyle behaviours, observed upon T2D diagnosis, lead, if at all, to a subsequent improvement in overall health outcomes in the elderly. In this study, I attempt to fill this gap and estimate the effect of T2D diagnosis on both physical and mental health outcomes for a sample of older adults in the UK. In a randomised evaluation, I would (randomly) diagnose some individuals with T2D, while leaving similar individuals undiagnosed, and estimate the difference in health outcomes in the two groups. In the absence of random assignment, however, retrieving such an estimate is very problematic because T2D diagnoses are assigned in a non-random manner, and generally targeted based on certain individual characteristics unobservable to the researcher. In this paper, I use a novel approach based on biomarkers data, collected through blood samples, to deal with the issue of nonrandom assignment. They key insight of the identification strategy is that General Practitioners (GP) diagnose their patients with T2D on the basis of an exact blood test, namely the fasting blood glucose (FBG) level, being above a pre-determined cut-off. Accordingly, I implement a regression discontinuity design (RDD) to estimate the impact of T2D diagnosis on both physical and mental health, by comparing individuals just below and just above the predetermined cut-off level of the FBG. The main source of data employed in the study is the English Longitudinal Study of Ageing (ELSA). This dataset is uniquely suited to answer the research question posed by this paper as it focusses on people aged 50 and above (and their partners), and as it contains detailed information on both physical and mental health, anthropometric measurements and various biomarkers data, collected through blood samples, which are at the core of this study. The RDD estimates reveal that health information, in the form of T2D, significantly matters for the health of the elderly. Specifically, relative to their counterparts, individuals diagnosed with T2D exhibit: (i) a significant lower level of overall and central obesity; (ii) significantly higher levels of self-reported body mobility; and, (iii) significantly higher scores on memory. However, I find no evidence that lifestyle recommendations have an effect on self-reported depression levels. In terms of magnitude of the estimated coefficients, the results appear to be strongest for men and for individuals below the age of 65. Importantly, the estimates are robust across a variety of specifications, including the inclusion (or exclusion) of different control variables and different functional form choices of the running variable. Additionally, I demonstrate that the results are unlikely to be driven by discontinuities in individual pre-diagnosis characteristics, or endogenous sorting around the threshold. The paper proceeds as follows. Section 2 presents the conceptual framework and outlines the quasi-experimental setting based on T2D diagnosis. Section 3 presents the data and reports summary statistics of the main variables of interest. In Section 4, the econometric methods employed in the analysis are explained. Section 5 presents the results, and Section 6 reports a series of robustness checks to confirm the validity and strength of the approach. The final Section provides a discussion of the results and concludes the paper. 2. The setting The key variable employed in the study is the person-specific fasting blood glucose (FBG) level, which is generally used as a measure of presence, or risk, of Type 2 Diabetes (henceforth, T2D). T2D is a metabolic disorder characterised by the body’s inability to use insulin properly, due to a combination of insulin resistance and beta-cells failure. As a result, the body cannot maintain normal blood glucose levels that, consequently, may become importantly

elevated. For this reason, the measurement of fasting blood glucose has been the conventional method for screening for T2D. While elevated blood glucose does not cause any immediate physical or mental symptom, if left untreated it can lead to heart disease, stroke and kidney problems in the long run. However, via adequate management and treatment much, if not all, of these complications can be avoided (Lim et al., 2001). According to the medical guidelines provided by the WHO (2006), the American Diabetes Association (ADA), as well as the European Association for the Study of Diabetes (EASD), if the FBG level is greater than or equal to 7 millimoles per litre (henceforth, mmol/L) then the individual is diagnosed with T2D.6 As suggested by the most recent National Institute for Health and Care Excellence (NICE) guidelines, the first line of treatment for T2Ddiagnosed subjects is the provision of a set of lifestyle recommendations to help patients manage their condition, which include: (i) promotion of healthy eating patterns low in fats and carbohydrates; (ii) reduction of energy intake to achieve substantial weight loss; (iii) reduction of sedentary time, including at least 150 min per week of moderate-intensity aerobic physical activity as well as engagement in resistance training at least twice a week; (iv) smoking and drinking cessation.7 Moreover, depending on the specific characteristics of the patients, GPs may assign T2D medications (e.g., metformin) to help lower blood glucose levels and restore insulin sensitivity, and may even prescribe bariatric surgery in cases of very elevated levels of BMI. 3. Data 3.1. Overview The study employs data from the English Longitudinal Study of Ageing (ELSA). The ELSA is a large-scale longitudinal panel study of people aged 50 and over, and their partners, living in private households in England. The original sample was drawn in 2002 from households that had previously responded to the Health Survey for England (HSE). Every two years, the sample has been interviewed to measure changes in their health status, economic conditions and social circumstances. The ELSA is employed not only because it contains detailed information of both physical and mental health but, most importantly, because at alternative waves a nurse visit is carried out in addition to the main interview.8 The nurse visit includes the collection of anthropometric measurements and various biological samples, collected through blood samples, which are the core of this study.9 Within three months of the nurse visit, study participants receive a letter with the result of each of the analyses conducted on the blood. If any result are out of the normal range participants are asked to contact their GP in the near future. Moreover, blood pressure, lung function and blood test results are

6 Since 2011, the WHO suggested that T2D may also be diagnosed by means of the haemoglobin A1c, which measures glycaemic control over the previous 3 months. Although the sample of interest was not affected by this change, I include this variable in the analysis to increase the precision of the point estimates. 7 Recent studies show significant improvements in glycaemic control for patients with type 2 diabetes that change to healthier lifestyles (Huang et al., 2016; Chen et al., 2015). Moreover, in a comprehensive review, Glechner et al. (2018) show that lifestyle changes may be an effective tool not only to manage but also to prevent T2D. 8 All core sample members have been offered a nurse visit at waves 2, 4 and 6. 9 All sample members who gave consent were eligible for a blood sample to be taken. Respondents were not asked to fast if they had diabetes and were on treatment or if they were considered to be malnourished or otherwise unfit to fast (this information was obtained from the interviewer). Respondents who were asked to fast were given guidelines about when and what they could eat based on their appointment time.

A. Gaggero / Economics and Human Biology 37 (2020) 100830

sent directly to the participants’ GP.10 Accordingly, to fully exploit the biomarkers in the data, this paper focuses on the waves in which the nurse visit was carried out, that is, waves 2, 4, and 6, collected in the years 2004, 2008, and 2012. 3.2. Key variables The ELSA provides a rich source of variables for the analysis. The first set of outcome variables measuring physical health include anthropometric measurements for both total as well as central obesity. I focus on obesity because it is associated with a striking increase in the risk of all-causes mortality, including heart disease, kidney disease and cancer (Jahangir et al., 2014; Kovesdy et al., 2017; Bhaskaran et al., 2014). Total obesity is captured by the widely used body mass index (BMI), which is a measure of a person’s weight relative to his/her height. One potential limitation of this index, however, is that it cannot differentiate between lean mass and fat mass and, consequently, may assign high values to individuals who are not overweight, but who are very muscular. To this end, in this study I also look at central obesity, as measured by waist circumference, which is a proxy for visceral fat in the body, and generally the underlying cause of increased risk of morbidity (Kuk et al., 2006). Self-assessed measurements for physical health are also considered. First, I consider the six-item activities of daily living (ADL) index developed by Katz et al. (1963). The ADL index measures the difficulties in performing tasks required for personal self-care and independent living in every-day life. The functional assessment is based on individuals’ responses (yes/no) at each wave to the six item questions asking them for (1) difficulties in dressing, (2) walking across a room, (3) bathing or showering, (4) eating, (5) getting out of bed, and (6) using the toilet. The overall score for each individual is calculated by summing across the itemspecific responses (Zaninotto and Falaschetti, 2011) and, therefore, ranges from zero, the most difficulties and worst physical health, to six, the least difficulties and best physical health. Additionally, participants in the ELSA were asked whether they had difficulty doing any of 10 activities that involved body mobility, such as (1) walking 100 yards, (2) sitting for two hours, (3) getting up from a chair after sitting long periods, (4) climbing several flights of stairs without resting, (5) climbing one flight of stairs without resting, (6) stooping, kneeling or crouching, (7) reaching or extending arms above shoulder level, (8) pushing or pulling large objects, (9) lifting or carrying weights over 10 pounds, and (10) picking up a 5p coin from a table. Similar to the above, for each item I generate a dummy variable which equals unity if the individual reports to have difficulty in a certain activity, and I create a body mobility index that ranges from zero to ten, where zero implies most difficulties, and ten implies least difficulties in mobility. Finally, I consider the set of outcome variables relating to mental health. In this study, mental health is measured in terms of learning and memory by means of the word-list learning test. In this test, participants are presented with ten common words and are asked to remember them. ELSA uses the word lists developed for health and retired studies (HRS), which comprise four different versions, so that different lists can be given to different members of the same household. The first member of the household to be tested is assigned a list at random by the computer, and where there is more than one member of the household in the ELSA sample, the remaining lists are also selected at random.11

10

http://www.elsa-project.ac.uk/uploads/elsa/docs_w2/nurse_leaflet.pdf. To ensure standardisation, the lists are presented by the computer, using a taped voice. 11

3

Lastly, I measure mental health in terms of depression, using the eight-item version of the CES-D scale developed by Radloff (1977). The CES-D is a depression screening test in which individuals were asked whether over the past week they had symptoms associated with depression, including: (1) if they felt depressed much of the time, (2) they felt everything they did was an effort, (3) their sleep was restless, (4) they were unhappy, (5) they felt lonely, (6) they enjoyed life, (7) they felt sad, and (8) if they could not get going. Similar to the above, responses were summed (Demakakos et al., 2010) for each individual to compute a total CES-D scale, ranging from zero, high depression and worst mental health status, to eight, no depression and best mental health.12 3.2.1. Descriptive analysis Table 1 shows summary statistics of the main variables of interest. In columns (1) and (2), I compare statistics of men and women, respectively, and in column (3) I report the p-value of a two-sided test that compares the means of the two groups. The table shows that the average age of the sample is 68, and that men in the survey are significantly more likely to be married and to have a higher degree classification. Women, on the other hand, are significantly more likely than men to smoke. The table shows that although there is no significant difference in BMI between men and women, men in the sample have a larger waist circumference than women. Following, men report significantly better scores in the self-assessed measurements of physical health than women, considering both the ADL and the body mobility indexes. Finally, women have a higher score in the wordlistening test, but are more likely than men to have symptoms associated with depression, as measured by the CES-D scale. Next, I report statistics of the key results obtained from the blood tests during the nurse visits. The table reports that men, on average, have slightly higher levels of fasting blood glucose (FBG) and of Haemoglobin A1c (HBA1c) than women. Further, the table reports that men are significantly more likely to be diagnosed with and taking medication for T2D. Finally, women are reported to have significantly higher levels of low-density lipoprotein (LDL) cholesterol (generally referred as bad cholesterol), but men have higher levels of both systolic and diastolic blood pressure. 4. Econometric methods In this study, I use a sharp Regression Discontinuity Design (RDD) to estimate the effect of T2D diagnosis on physical and mental health. This type of design was first introduced by Thistlethwaite and Campbell (1960) and then formalised by Hahn et al. (2001), who derived the necessary conditions for the identification of causal effects. Despite being underutilised in health economics and epidemiology (Moscoe et al., 2015), RDD is becoming increasingly popular in empirical studies given the assumptions needed for the identification of causal effects are quite weak. The defining feature of this class of models is that the probability of receiving the treatment changes discontinuously as a function of an assignment variable, denoted as Z i;t ; being above or below a certain cut-off point, denoted as  z0 .13 The underlying idea of an RDD is that, as in a randomised experiment, for individuals just above and below the pre-identified cut-off point, assignment

12 The CES-D scale has been validated in older populations and displays strong psychometric properties (Zivin et al., 2010; Lyness et al., 1997). 13 There are two types of RDD: the sharp and fuzzy design. In the sharp design, treatment status depends deterministically on the running variable being above or below the cut-off level. In contrast, in the fuzzy design the probability of receiving the treatment is known to be discontinuous in the cut-off point, but not in a deterministic manner. While the sharp design is the preferred model, I also employ a fuzzy design to check the robustness of the results.

4

A. Gaggero / Economics and Human Biology 37 (2020) 100830

Table 1 Summary Statistics.

Attributes: Years of Age Married [0,1] Family Size Higher Education [0,1] Log-equivalised HH Income Smoking [0,1] Outcome Variables: Body Mass Index (Kg/m2) Waist Circumference (cm) ADL Index [0,6] Body Mobility Index [0,10] Word-Listing [0,10] CES-D Scale [0,8] Biomarkers: FBG (mmol/L) Haemoglobin A1c (%) T2D Diagnosis [0,1] T2D Treatment [0,1] LDL Cholesterol (mmol/L) Systolic Pressure (mm Hg) Diastolic Pressure (mm Hg) Observations

(1) Men

(2) Women

(3) p -value

68.51 (8.192) 0.77 (0.419) 2.05 (0.762) 0.45 (0.498) 5.66 (0.661) 0.07 (0.248)

68.53 (8.435) 0.60 (0.491) 1.85 (0.749) 0.28 (0.448) 5.56 (0.680) 0.09 (0.281)

0.905

27.95 (4.333) 101.62 (11.513) 5.77 (0.699) 8.72 (2.046) 5.76 (1.667) 6.26 (1.256)

28.04 (5.485) 91.57 (12.830) 5.73 (0.759) 7.99 (2.397) 6.17 (1.677) 5.80 (1.554)

0.426

4.99 (0.874) 5.73 (0.730) 0.10 (0.298) 0.05 (0.219) 3.28 (0.998) 134.41 (16.346) 75.81 (10.525) 3293

4.88 (0.693) 5.70 (0.596) 0.07 (0.251) 0.03 (0.178) 3.60 (1.027) 132.00 (18.164) 74.62 (10.255) 4099

0.000

0.000 0.000 0.000 0.000

5. Results

0.001

0.000 0.030 0.000 0.000 0.000

0.088 0.000 0.000 0.000 0.000 0.000 7392

Note: The table reports summary statistics of the variables of interest, comparing men and women. Standard deviations in parentheses.Source: English Longitudinal Study of Ageing (ELSA).

to treatment is as good as random and, therefore, omitted variable bias disappears. Formally, let Z i;t be the running variable which identifies the fasting blood glucose level for individual i at time t, and let the cutoff point of interest be  z0 ¼ 7. Further, let us denote the indicator
 function  Di;t ¼ 1 Z i;t    z0 , which takes the value of unity for subjects with fasting blood glucose levels above the pre-determined cut-off and, consequently, diagnosed with T2D, such that: 
   f 1
Z i;t ;  if  Z i;t < z0 ð1Þ Pr Di;t ¼ 1 ¼ ; f 0 Z i;t ;  if  Z i;t  z0 where, due to the discontinuity at the cut-off point,
 f 1 Z i;t 6¼  f 0 ðZ i;t Þ. In the spirit of Hahn et al. (2001), I estimate a version of Eq. (2) on a number of physical and mental health outcomes as follows: 0

Y i;tþs ¼ a þ bDi;t þ dZ i;t þ  X i;t g þ ei;t ;  t ¼ 2004;  2008: Y i;tþs denotes the outcome of interest for individual i at time t þ s (where s ¼ 4); Di;t and  Z i;t are respectively the T2D diagnosis binary variable and the running variable as described above. b is the main term of interest, as it measures the effect of T2D diagnosis 0

size, marital status, household income, and smoking behaviour, as well as a set of baseline biomarkers, including Haemoglobin A1c, LDL cholesterol, systolic and diastolic blood pressure, which are also key markers considered by the GP when diagnosing T2D. To take into account the medical history of the subjects, I also include a set of dummy variables that identify subjects’ pre-existing medical conditions. Finally, ei;t is a random error term. Standard errors are clustered at the individual level because individuals appear in the regressions in multiple waves, but results are insensitive to different specifications.14

on the outcomes of interest. X i;t includes a set of baseline respondents’ attributes, including age, gender, education, family

In Fig. 1, I begin to investigate the effect of T2D diagnosis on physical and mental health, graphically, by plotting a series of (unconditional) local polynomial smoothing (LPS) regressions of the main outcomes of interest at time t þ s as a function of the baseline fasting blood glucose (FBG) at time t.15 Fig. 1 shows that, after receiving T2D diagnosis, individuals with FBG just above the threshold report a jump in the levels of overall and central obesity, and also a jump with respect to both ADL and body mobility indexes. Finally, the figure does not report a discontinuity in word listening and self-reported depression test. In what follows, I test the significance of these findings in a regression framework, as specified in Eq. (2), when controlling for a number of other possible confounding factors. Table 2 reports RDD estimates of the effect of T2D diagnosis on health outcomes for the sample of interest. Columns (1) and (2) imply that four years after receiving T2D diagnosis individuals report a statistically significant lower score in their BMI level (over 2 points), and around 5.5 cm less in their waist circumference than their counterparts. In columns (3) and (4), the table reports that individuals diagnosed with T2D exhibited statistically significant better scores in the ADL and in the body mobility indexes, implying they reported encountering fewer difficulties in performing activities of daily living, as well as performing physical movements. Specifically, the estimated coefficients imply that four years after receiving the diagnosis, relative to their counterpart, individuals reported a 0.19 and 0.70 increase in the scores of ADL and body mobility indexes, respectively. Finally, columns (5) and (6) display that individuals diagnosed with T2D exhibit a 0.5 higher memory and learning score than their counterparts, as measured by the word-listening test, but insignificant effects were found in the reported depression level.16

14 To test the robustness of the results, I employ a fuzzy RDD as an alternative approach. Drawing from the above notation, let SRDi;tþs take the value of unity for patients with self-reported T2D. Then, I estimate the following model: 0 Y i;tþs ¼ m0 þ m1 SRDi;tþs þ m2 Z i;t þ  X i;t m3 þ ei;t ; Where SRDi;tþs ; is instrumented by the indicator Di;t . The results of this alternative method are reported in Table A6 in the Appendix and are in line with the overall findings of the paper. 15 Due to the small sample size on the right-hand side of the cut-off, the unconditional LPS regressions report very big standard errors. We address this in a regression framework by including a full set of covariates, as explained in Section 4, and by clustering the standard errors at the individual level. 16 Additionally, in table A7 I explore in a descriptive manner whether improvements in subjects’ self-assessed physical and mental health outcomes are driven by the observed reduction in obesity levels. To do so, I use an instrumental variable (IV) approach that investigates the relationship between reduced BMI and physical and mental health. Specifically, first, I estimate the 0 following equation: Y i;tþs ¼ p0 þ p1 BMIi;tþs þ p2 Zi;t þ  X i;t p3 þ ei;t ; In which the term BMIi;tþs is the predicted reduced BMI of individual iat time t þ s, as predicted by the following first-stage equation: 0 BMIi;tþs ¼ v0 þ v1 Di;t þ v2 Z i;t þ  X i;t v2 þ ui;t . It is worth mentioning that the estimated coefficient p1 cannot be interpreted as causal because the implied exclusion restriction by this approach would not be satisfied.

A. Gaggero / Economics and Human Biology 37 (2020) 100830

5

Fig. 1. Outcomes of Interest at the Cut-off. Note: The figure shows local polynomial estimates of the discontinuity on the outcomes of interest at time t + s, as a function of the fasting blood glucose (FBG) at time t.

Overall, the table shows there are other variables that significantly affect the outcomes of interest. As expected, variables such as education and household income have a strong and significant explanatory power for all outcomes. Subjects who completed higher education and subjects that belong to richer households, report lower obesity levels and better scores in terms of selfassessed physical and mental health. Smoking behaviour also has a strong effect on the outcomes. Interestingly, individuals who were reported to smoke have lower BMI and waist-circumference levels, but report statistically significant worse scores in the remaining outcomes that identify healthy ageing. To conclude, Table 3 reports heterogeneous effects of T2D diagnosis by disaggregating the sample in different categories. The results show that, relatively to women, men report greater improvements and that, while T2D diagnosis has positive effects at all ages, in terms of the magnitude of the estimated

coefficients those below the age of 65 are those that benefitted. To sum up, the findings provide strong evidence that health information, in the form of T2D diagnosis, may have important effects on the health of the elderly. In the section that follows, I present a series of tests and checks to confirm the validity of these findings.17

17 In Table A1 in Appendix, I also report estimated effects of T2D diagnosis on a number of respondents’ biomarkers. The results show that individuals diagnosed with T2D report lower levels of blood pressure (systolic), higher levels of the highdensity lipoprotein (HDL) cholesterol (often referred as good cholesterol), and lower levels of triglycerides (a measure of fat in the blood often associated with the risk of heart disease). Further, in Tables A2–A4, I report the effect of T2D diagnosis on the self-assessed outcomes (ADL index, body mobility index, and CES-depression scale) by splitting the indexes into their individual components.

6

A. Gaggero / Economics and Human Biology 37 (2020) 100830

Table 2 RD Estimates of Physical and Mental Health.

Physical Health Anthropometrics

T2D Diagnosis [0, 1] Running Variable: FBG (mmol/L) Attributes: Female [0,1] Years of Age Years of Age2 Higher Education [0,1] Family Size Married [0,1] Log-equivalised HH Income Smoking [0,1] Biomarkers: Haemoglobin A1c (%) LDL Cholesterol (mmol/L) Systolic Pressure (mm Hg) Diastolic Pressure (mm Hg) Taking Medications for: T2D Treatment [0,1] High Cholesterol [0,1] Hypertension [0,1] Time dummy Constant Observations

Physical Health Self-Assessed

Mental Health

(1) Body Mass Index

(2) Waist Circumference

(3) ADL Index

(4) Body Mobility Index

(5) Word Listening Test

(6) CES – D Scale

2.149*** (0.568)

5.521*** (1.307)

0.193** (0.090)

0.702*** (0.271)

0.427* (0.236)

0.245 (0.213)

0.361*** (0.098)

0.970*** (0.264)

0.030 (0.020)

0.038 (0.059)

0.049 (0.038)

0.001 (0.036)

0.308* (0.157) 0.341** (0.153) 0.003** (0.001) 0.808*** (0.164) 0.237* (0.124) 0.001 (0.187) 0.143** (0.066) 0.930*** (0.272)

9.710*** (0.383) 0.614 (0.380) 0.005* (0.003) 2.122*** (0.400) 0.204 (0.261) 0.076 (0.442) 0.427** (0.199) 0.898 (0.680)

0.017 (0.021) 0.047* (0.027) 0.000** (0.000) 0.056*** (0.021) 0.067** (0.028) 0.139*** (0.036) 0.051*** (0.012) 0.093** (0.045)

0.617*** (0.067) 0.158** (0.074) 0.002*** (0.001) 0.374*** (0.067) 0.113* (0.059) 0.291*** (0.091) 0.225*** (0.041) 0.539*** (0.133)

0.536*** (0.047) 0.166*** (0.057) 0.002*** (0.000) 0.460*** (0.050) 0.010 (0.035) 0.016 (0.059) 0.188*** (0.032) 0.212** (0.083)

0.379*** (0.043) 0.069 (0.054) 0.001 (0.000) 0.127*** (0.044) 0.036 (0.038) 0.391*** (0.059) 0.140*** (0.031) 0.225** (0.091)

0.866*** (0.162) 0.170** (0.068) 0.010** (0.005) 0.039*** (0.008)

2.253*** (0.401) 0.395** (0.166) 0.035*** (0.012) 0.069*** (0.020)

0.019 (0.026) 0.007 (0.011) 0.000 (0.001) 0.001 (0.001)

0.186** (0.081) 0.047 (0.032) 0.001 (0.002) 0.002 (0.004)

0.068 (0.055) 0.047** (0.024) 0.004** (0.002) 0.006** (0.003)

0.095* (0.053) 0.010 (0.022) 0.000 (0.002) 0.002 (0.003)

2.397*** (0.841) 0.273* (0.157) 0.807*** (0.141) 0.212** (0.095) 8.029 (5.153) 4970

6.530*** (1.544) 0.692 (0.437) 2.237*** (0.364) 1.253*** (0.239) 56.991*** (13.221) 4970

0.120 (0.096) 0.039 (0.037) 0.090*** (0.028) 0.031 (0.020) 4.311*** (0.949) 4970

0.836*** (0.308) 0.193* (0.099) 0.272*** (0.081) 0.114** (0.052) 4.904* (2.552) 4970

0.023 (0.185) 0.163** (0.078) 0.017 (0.058) 0.247*** (0.044) 1.508 (1.948) 4970

0.410* (0.221) 0.213*** (0.071) 0.114** (0.055) 0.109*** (0.041) 3.794** (1.860) 4970

Note: The table presents regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on the outcomes of interest. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

Table 3 RD Estimates of Physical and Mental Health, Heterogeneous Effects. Physical Health Anthropometrics

Men: T2D Diagnosis [0,1] Observations Women: T2D Diagnosis [0, 1] Observations Age  65: T2D Diagnosis [0, 1] Observations Age<65: T2D Diagnosis [0,1] Observations

Physical Health Self-Assessed

Mental Health

(1) Body Mass Index

(2) Waist Circumference

(3) ADLIndex

(4) Body Mobility Index

(5) Word Listening Test

(6) CES - DScale

2.403*** (0.626) 2213

5.038*** (1.640) 2213

0.187 (0.115) 2213

0.629* (0.332) 2213

0.714** (0.338) 2213

0.085 (0.235) 2213

1.434 (0.895) 2757

4.764** (1.960) 2757

0.144 (0.143) 2757

0.504 (0.441) 2757

0.130 (0.320) 2757

0.295 (0.380) 2757

1.706** (0.719) 2996

4.108*** (1.544) 2996

0.159 (0.098) 2996

0.448 (0.323) 2996

0.706** (0.282) 2996

0.173 (0.254) 2996

5.138*** (1.458) 1974

13.187*** (2.658) 1974

0.465** (0.207) 1974

1.717*** (0.534) 1974

0.270 (0.418) 1974

0.788** (0.376) 1974

Note: The table presents regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on the outcomes of interest. Each coefficient in the table is from a separate regression. Results are conditional on a vector of covariates, as reported in Table 2. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

A. Gaggero / Economics and Human Biology 37 (2020) 100830

7

Fig. 2. RDD Continuity Test. Note: The figure shows local polynomial estimates of a number of control variables, as a function of the fasting blood glucose (FBG). Source: English Longitudinal Study of Ageing (ELSA).

6. Robustness checks

Borrowing from the jargon of the program evaluation literature, let

The main assumption that needs to be satisfied for the above identification strategy to produce unbiased estimates is continuity.

fY 1i ;  Y 0i g be the potential outcome for individual i in case of either receiving or not receiving T2D diagnosis, respectively. Then, the continuity assumption that needs to be satisfied for the validity of a

8

A. Gaggero / Economics and Human Biology 37 (2020) 100830

Table 4 RD Estimates, Donut-hole Approach. Physical Health Anthropometrics

T2D Diagnosis [0, 1] Running Variable: FBG (mmol/L) Attributes: Female [0,1] Years of Age Years of Age2 Higher Education [0,1] Family Size Married [0,1] Log-equivalised HH Income Smoking [0,1] Biomarkers: Haemoglobin A1c (%) LDL Cholesterol (mmol/L) Systolic Pressure (mm Hg) Diastolic Pressure (mm Hg) Taking Medications for: T2D Treatment [0,1] High Cholesterol [0,1] Hypertension [0,1] Time dummy Constant Observations

Physical Health Self-Assessed

Mental Health

(1) Body Mass Index

(2) Waist Circumference

(3) ADL Index

(4) Body Mobility Index

(5) Word Listening Test

(6) CES – D Scale

3.060*** (0.803)

7.566*** (1.897)

0.249* (0.128)

0.676* (0.378)

0.545* (0.298)

0.427 (0.261)

0.378*** (0.101)

1.004*** (0.270)

0.035 (0.022)

0.027 (0.062)

0.050 (0.039)

0.001 (0.038)

0.270* (0.158) 0.340** (0.154) 0.003** (0.001) 0.813*** (0.164) 0.250** (0.125) 0.050 (0.186) 0.139** (0.066) 0.985*** (0.271)

9.759*** (0.385) 0.603 (0.383) 0.005* (0.003) 2.152*** (0.402) 0.227 (0.262) 0.122 (0.444) 0.420** (0.200) 1.085 (0.665)

0.017 (0.021) 0.047* (0.027) 0.000* (0.000) 0.053** (0.021) 0.066** (0.028) 0.136*** (0.036) 0.050*** (0.012) 0.088** (0.045)

0.606*** (0.068) 0.147* (0.075) 0.001*** (0.001) 0.358*** (0.067) 0.118** (0.060) 0.284*** (0.091) 0.225*** (0.041) 0.505*** (0.132)

0.528*** (0.047) 0.170*** (0.057) 0.002*** (0.000) 0.455*** (0.050) 0.007 (0.035) 0.021 (0.059) 0.190*** (0.032) 0.216** (0.084)

0.371*** (0.043) 0.072 (0.054) 0.001 (0.000) 0.127*** (0.044) 0.044 (0.039) 0.382*** (0.059) 0.138*** (0.031) 0.223** (0.092)

0.893*** (0.168) 0.173** (0.068) 0.011** (0.005) 0.039*** (0.008)

2.352*** (0.424) 0.385** (0.167) 0.039*** (0.012) 0.067*** (0.020)

0.030 (0.028) 0.006 (0.011) 0.000 (0.001) 0.001 (0.002)

0.181** (0.085) 0.038 (0.033) 0.001 (0.003) 0.002 (0.004)

0.079 (0.058) 0.046* (0.024) 0.004** (0.002) 0.006** (0.003)

0.101* (0.055) 0.012 (0.023) 0.001 (0.002) 0.002 (0.003)

2.591*** (0.957) 0.253 (0.159) 0.769*** (0.143) 0.218** (0.095) 10.300** (5.163) 4899

7.114*** (1.757) 0.733* (0.441) 2.193*** (0.368) 1.289*** (0.242) 63.266*** (13.387) 4899

0.135 (0.102) 0.042 (0.038) 0.088*** (0.028) 0.036* (0.020) 4.162*** (0.963) 4899

0.906** (0.352) 0.215** (0.100) 0.263*** (0.082) 0.111** (0.053) 5.064* (2.623) 4899

0.055 (0.200) 0.184** (0.079) 0.011 (0.059) 0.245*** (0.044) 1.091 (1.975) 4899

0.465* (0.242) 0.220*** (0.072) 0.107* (0.056) 0.099** (0.041) 3.742** (1.886) 4899

Note: The table presents regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on the outcomes of interest, employing the donut-hole approach with excluded 0.5 mmol/L around the cut-off. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

(sharp) RDD can be formally written as follows: h  h   0  E Y 0i Z i ¼ zþ 0    E Y i Z i ¼ z0  ¼ 0;

ð3Þ

 where zþ 0 and z0 represent, respectively, people just above and below the cut-off point, z0 . In this case, the continuity assumption entails that the people just above and just below the pre-identified cut-off level of fasting blood glucose are identical in every aspect, both in terms of observable and unobservable characteristics, but only differ in the probability of being assigned lifestyle recommendations. A direct way to assess this assumption is to examine if baseline variables change discontinuously around the cut-off point. Accordingly, in Fig. 2, I plot a set of local polynomial smoothing (LPS) regressions of baseline variables including: (a) log-equivalised household income, (b) education level; (c) bodyweight; (d) body height; (e) haemoglobin A1c; (f) LDL cholesterol; (g) systolic; and, (h) diastolic blood pressure. The figure does not show any statistically significant discontinuity at the cut-off point for any of these variables.18

18

Analogous plots for other covariates show very similar patterns.

A potential threat for the validity of the identification could arise in case of manipulation of the forcing variable. As McCrary (2008) argues, if individuals can manipulate the score of Z i;t , in order to receive or not receive the treatment, then the continuity assumptions will not be satisfied. In my case, this may have occurred if study participants changed their behaviour in anticipation of the nurse visit. Although visual inspection (Fig. A1) reveals no suspect discontinuity in the distribution of the running variable,19 one way to account for the potential endogenous sorting around the running variable is the so called “donut hole” approach, suggested by Barreca et al. (2016). The main idea behind this approach is that units closest to the cut-off are those most likely to have engaged in manipulation. Consequently, excluding such units from the analysis would eliminate this potential concern. Accordingly, Table 4 reports RDD estimates of T2D diagnosis on physical and mental health outcomes, excluding subjects with FBG values within 0.50 mmol/L around

19 The McCrary density test confirms that I cannot reject the hypothesis that there is no shift in the discontinuity. The estimated log difference in the height of the density is 0.90 with a standard error or of 0.49.

A. Gaggero / Economics and Human Biology 37 (2020) 100830

9

Table 5 Non-Parametric RD Estimates.

T2D Diagnosis [0,1] Running Variable: FBG (mmol/L) Attributes: Female [0,1] Years of Age Years of Age2 Higher Education [0,1] Family Size Married [0,1] Log-equivalised HH Income Smoking [0,1] Biomarkers: Haemoglobin A1c (%) LDL Cholesterol (mmol/L) Systolic Pressure (mm Hg) Diastolic Pressure (mm Hg) Taking Medications for: T2D Treatment [0,1] High Cholesterol [0,1] Hypertension [0,1] Time dummy Constant Observations

Physical Health Anthropometrics

Physical Health Self-Assessed

Mental Health

(1) Body Mass Index

(2) Waist Circumference

(3) ADL Index

(4) Body Mobility Index

(5) Word Listening Test

(6) CES – D Scale

2.124*** (0.792)

3.921** (1.656)

0.196* (0.111)

0.973*** (0.327)

0.516* (0.296)

0.387 (0.272)

0.861*** (0.249)

1.799*** (0.615)

0.070 (0.045)

0.237* (0.124)

0.026 (0.091)

0.145 (0.089)

0.800*** (0.249) 0.380 (0.275) 0.003* (0.002) 1.016*** (0.252) 0.133 (0.199) 0.387 (0.305) 0.169 (0.128) 0.863** (0.405)

8.392*** (0.595) 0.782 (0.629) 0.006 (0.005) 2.682*** (0.609) 0.184 (0.474) 0.793 (0.718) 0.496 (0.319) 0.388 (1.136)

0.014 (0.036) 0.060 (0.044) 0.000 (0.000) 0.063* (0.036) 0.105** (0.052) 0.171*** (0.059) 0.057*** (0.021) 0.048 (0.062)

0.758*** (0.108) 0.220* (0.118) 0.002** (0.001) 0.375*** (0.107) 0.086 (0.105) 0.302** (0.149) 0.329*** (0.066) 0.489** (0.238)

0.556*** (0.073) 0.091 (0.092) 0.001* (0.001) 0.393*** (0.076) 0.067 (0.054) 0.128 (0.089) 0.238*** (0.046) 0.125 (0.137)

0.492*** (0.069) 0.170** (0.086) 0.001** (0.001) 0.171** (0.068) 0.019 (0.067) 0.392*** (0.096) 0.135*** (0.047) 0.183 (0.165)

0.881*** (0.282) 0.499*** (0.119) 0.012 (0.009) 0.051*** (0.015)

2.034*** (0.638) 1.079*** (0.260) 0.020 (0.019) 0.120*** (0.032)

0.032 (0.044) 0.019 (0.018) 0.001 (0.002) 0.002 (0.003)

0.078 (0.123) 0.176*** (0.049) 0.001 (0.004) 0.010 (0.007)

0.132 (0.082) 0.029 (0.037) 0.001 (0.003) 0.007 (0.005)

0.059 (0.079) 0.026 (0.034) 0.002 (0.003) 0.003 (0.005)

1.517** (0.757) 0.125 (0.266) 0.656*** (0.230) 0.217 (0.189) 4.964 (9.479) 1927

4.430*** (1.515) 0.413 (0.661) 2.015*** (0.523) 1.085** (0.435) 51.207** (22.219) 1927

0.100 (0.132) 0.003 (0.059) 0.088** (0.041) 0.012 (0.034) 3.544** (1.568) 1927

0.855** (0.358) 0.079 (0.161) 0.361*** (0.122) 0.137 (0.087) 1.095 (4.189) 1927

0.090 (0.231) 0.067 (0.111) 0.034 (0.091) 0.282*** (0.073) 3.754 (3.239) 1927

0.304 (0.266) 0.080 (0.108) 0.107 (0.086) 0.128* (0.068) 0.674 (3.044) 1927

Note: The table presents non-parametric regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on the outcomes of interest, at a window of [+2,-2] mmol/L around the cut-off. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

Table 6 RD Estimates, with Higher Order Polynomials of the Running Variable.

Quadratic: T2D Diagnosis [0, 1]

Cubic: T2D Diagnosis [0,1] Observations Quartic: T2D Diagnosis [0,1] Observations

Physical Health Anthropometrics

Physical Health Self-Assessed

Mental Health

(1) Body Mass Index

(2) Waist Circumference

(3) ADL Index

(4) Body Mobility Index

(5) Word Listening Test

(6) CES – D Scale

1.571** (0.667) 4970

3.701* (1.379) 4970

0.179** (0.091) 4970

0.692** (0.311) 4970

0.457* (0.243) 4970

0.162 (0.219) 4970

1.351* (0.729) 4970

3.616** (1.597) 4970

0.125 (0.118) 4970

0.902** (0.356) 4970

0.517* (0.300) 4970

0.266 (0.283) 4970

1.288* (0.713) 4970

3.547** (1.568) 4970

0.127 (0.119) 4970

0.932*** (0.349) 4970

0.526* (0.300) 4970

0.283 (0.285) 4970

Note: The table presents regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on the outcomes of interest, with higher order polynomials of the running variable. Each coefficient in the table is from a separate regression. Results are conditional on a vector of covariates, as reported in Table 2. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

10

A. Gaggero / Economics and Human Biology 37 (2020) 100830

the pre-identified cut-off. As the table shows, the results remain virtually unchanged. 20 Notably, the findings are robust across a variety of specifications. These estimates are robust to the inclusion/exclusion of various control variables and different functional form choices. Moreover, Table 5 shows the robustness of the results by performing a non-parametric RDD, that is, an approach in which Eq. (2) is estimated only for a sample of subjects in arbitrarily small neighbourhoods around the cut-off point. I report estimated coefficients when using a window of [+2,-2] mmol/L around the cut-off, bur results are robust when using different windows. Additionally, Table 6 presents RDD estimates of the diagnosis with different function forms of the running variable. Although the results remain consistent across the different specification, when including third and fourth degree polynomials of the running variable, the estimated coefficient of interest loses its statistical significance for the ADL index.21 7. Conclusion Current medical guidelines cause a discontinuity in the probability individuals receive type 2 diabetes diagnosis (T2D) as a function of exact blood test results. This feature is exploited in a regression discontinuity design (RDD) framework to identify the causal effects of T2D diagnosis on physical and mental health for a sample of individuals aged 50 and above. Employing the English Longitudinal Study of Ageing, I find compelling evidence that health information, in the form of T2D diagnosis, significantly matters for the health of the elderly. In particular, upon receiving T2D diagnosis individuals reported considerably lower levels of overall and central obesity, less difficulties in body mobility, and performed better in the learning memory test. Importantly, these findings are robust to a battery of tests that confirm the strength and validity of the approach. One potential shortcoming of the current approach is that, as explained in Section 2, upon receiving T2D diagnosis, individuals may simultaneously be assigned to a number of lifestyle recommendations (e.g., diet, exercise, smoking and drinking behaviour) as well as specific medications and/or treatments to help improve glycaemic control. Regrettably, the current data does not permit to estimate the specific effect of the different components.

20 Additionally, the findings are robust when implementing the treatment effect derivative (TED) of the estimated RDD, constructed by Dong and Lewbel (2015), which tests for the stability of the RDD estimates (Cerulli et al., 2017). 21 In Table A5 of the Appendix, I report a series of falsification tests in which I present RDD estimates for a number different cut-offs just before the predetermined value, that is, with z0 ranging from 6 to 6.9 mmol/L. The table shows that, with the only exception of the 6.9 cut-off value, virtually none of the estimated coefficients is statistically significant at the different cut-offs value.

This paper may have important policy implications. Firstly, to the extent that people respond to health information, as in the case of T2D diagnosis, the results of this paper suggests that health information and health screening programs may be an effective tool policymakers might use as a pre-emptive measure to promote healthy behaviours. Further, to the extent to which jobs and occupations require the performance of both physical and mental activities, this study suggests that health information and screening programs could also have important benefits in the productivity of the elderly and, potentially, prolong their participation in the labour force beyond retirement age. As this is one of the first studies to evaluate the effect of a T2D diagnosis exploiting biomarkers data in an RDD framework, more research is needed to verify whether the findings can be observed in other contexts. In particular, as highlighted from Slade (2012), it is possible that the estimated effects may gradually disappear with time. Accordingly, future work should investigate the long term effects of the diagnosis. Moreover, further studies should examine whether, and the extent to which, the observed benefits on physical and mental health outcomes can spill over to other family members and in the labour market. CRediT authorship contribution statement Alessio Gaggero: . Acknowledgements I thank the editor, Susan Averett, and two anonymous referees for comments that have helped improve the article. I thank M.D. Chiara Serafini for constructive discussions on type 2 diabetes. I thank Denni Tommasi, Simona Demel, Sourafel Girma, Eugenio Zucchelli, and Dolores Jimenez for comments and feedback at the early stages of the paper. Finally, I am very grateful to Lina Song, Getinet Haile, Bing Liu, Jing Zhang, Mo Tian, Arijit Mukherjee, David Morris, and other seminar participants of the Nottingham University Business School and participants of the XXXIX Jornadas de Economia de la Salud for helpful and constructive comments. All errors are mine. Appendix A

Fig. A1. Manipulation of RunningVariable. Note: The gure reports evidence of no manipulation around the cut-off. Source: English Longitudinal Study of Ageing (ELSA).

A. Gaggero / Economics and Human Biology 37 (2020) 100830

11

Table A1 RD Estimates of Other Biomarkers.

T2D Diagnosis [0,1] Observations

Glycemic Control:

Blood Pressure:

(1) Fasting Glucose

(2) Hemoglobin A1c

(3) Systolic Pressure

(4) Diastolic Pressure

(5) Total Cholesterol

(6) LDL Cholesterol

(7) HLD Cholesterol

(8) Triglycerides Level

(9) C-Reactive Protein

0.514

0.213

5.761**

1.165

0.086

0.106

0.157**

0.260**

0.485

(0.540) 2778

(0.146) 4204

(2.393) 4737

(1.375) 4737

(0.158) 4260

(0.140) 4227

(0.075) 4260

(0.125) 4260

(1.362) 4260

Blood Lipids:

Inflammation:

Note: The table presents regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on different biomarkers. Each coefficient in the table is from a separate regression. Results are conditional on a vector of covariates, as reported in Table 2. Robust standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

Table A2 RD Estimates of Activities of Daily Living.

T2D Diagnosis [0,1] Observations

(1) Dressing

(2) Walking Across a Room

(3) Bathing or Showering

(4) Cutting up Food

(5) Using the Toilet

(6) Getting In and Out of Bed

0.084* (0.044) 4970

0.022** (0.010) 4970

0.025 (0.040) 4970

0.022** (0.011) 4970

0.002 (0.022) 4970

0.053** (0.025) 4970

Note: The table presents regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on the ADL specific components. Each coefficient in the table is from a separate regression. Results are conditional on a vector of covariates, as reported in Table 2. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

Table A3 RD Estimates of Body Mobility.

T2D Diagnosis [0,1] Observations

(1) Walk 100 Yards

(2) Sit 2 Hours

(3) Get up from a Chair

(4) Climb Several Stairs

(5) Climb One Stair

(6) Kneel or Crouch

(7) Extend Arms

(8) Push or Pull Objects

(9) Lift Weights

(10) Pick up 5p Coin

0.058 (0.047) 4970

0.097** (0.042) 4970

0.108* (0.062) 4970

0.041 (0.063) 4970

0.139*** (0.049) 4970

0.137** (0.065) 4970

0.030 (0.043) 4970

0.079 (0.051) 4970

0.038 (0.058) 4970

0.027 (0.026) 4970

Note: The table presents regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on body mobility specific components. Each coefficient in the table is from a separate regression. Results are conditional on a vector of covariates, as reported in Table 2. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

Table A4 RD Estimates of CES- Depression Scale.

T2D Diagnosis [0,1] Observations

(1) Depressed Much of the Time

(2) Everything was Effort

(3) Sleep is Restless

(4) Unhappy

(5) Lonely

(6) Enjoyed Life

(7) Sad

(8) Could not Get going

0.047 (0.051) 4970

0.039 (0.054) 4970

0.088 (0.064) 4970

0.027 (0.045) 4970

0.013 (0.041) 4970

0.002 (0.042) 4970

0.034 (0.056) 4970

0.033 (0.058) 4970

Note: The table presents regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on the CES - depression scale specific components. Each coefficient in the table is from a separate regression. Results are conditional on a vector of covariates, as reported in Table 2. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

12

A. Gaggero / Economics and Human Biology 37 (2020) 100830

Table A5 RD Estimates of Physical and Mental Health, at different cut-offs.

Using 6.0 cut-off Using 6.1 cut-off Using 6.2 cut-off Using 6.3 cut-off Using 6.4 cut-off Using 6.5 cut-off Using 6.6 cut-off Using 6.7 cut-off Using 6.8 cut-off Using 6.9 cut-off

Physical Health Anthropometrics

Physical Health Self-Assessed

Mental Health

(1) Body Mass Index

(2) Waist Circumference

(3) ADL Index

(4) Body Mobility Index

(5) Word Listening Test

(6) CES – D Scale

1.164** (0.592) 0.123 (0.310) 0.128 (0.345) 0.128 (0.345) 0.230 (0.371) 0.316 (0.363) 0.575 (0.460) 0.793 (0.557) 0.793 (0.557) 1.492*** (0.562)

1.092 (0.775) 0.141 (0.868) 0.053 (0.944) 0.053 (0.944) 1.033 (1.026) 1.035 (1.070) 1.855 (1.282) 1.899 (1.414) 1.899 (1.414) 3.797*** (1.278)

0.025 (0.070) 0.023 (0.081) 0.006 (0.093) 0.006 (0.093) 0.110 (0.082) 0.160* (0.082) 0.160* (0.085) 0.094 (0.083) 0.094 (0.083) 0.152* (0.089)

0.225 (0.176) 0.212 (0.184) 0.277 (0.203) 0.277 (0.203) 0.035 (0.194) 0.155 (0.209) 0.304 (0.232) 0.246 (0.245) 0.246 (0.245) 0.540** (0.250)

0.093 (0.127) 0.012 (0.139) 0.005 (0.152) 0.005 (0.152) 0.003 (0.162) 0.065 (0.165) 0.233 (0.199) 0.215 (0.207) 0.215 (0.207) 0.309 (0.231)

0.295** (0.124) 0.259* (0.141) 0.176 (0.152) 0.176 (0.152) 0.066 (0.159) 0.011 (0.166) 0.064 (0.177) 0.115 (0.182) 0.115 (0.182) 0.180 (0.206)

Note: The table presents regression discontinuity (RD) estimates of the effect of type 2 diabetes diagnosis on the outcomes of interest. Each coefficient in the table is from a separate regression. Results are conditional on a vector of covariates, as reported in Table 2. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

Table A6 Fuzzy RD Estimates of Physical and Mental Health.

Self-Reported T2D [0, 1] Running Variable: FBG (mmol/L) Attributes: Female [0,1] Years of Age Years of Age2 Higher Education [0,1] Family Size Married [0,1] Log-equivalised HH Income Smoking [0,1] Biomarkers: Haemoglobin A1c (%) LDL Cholesterol (mmol/L) Systolic Pressure (mm Hg) Diastolic Pressure (mm Hg) Time dummy Constant Observations

Physical Health Anthropometrics

Physical Health Self-Assessed

Mental Health

(1) Body Mass Index

(2) Waist Circumference

(3) ADL Index

(4) Body Mobility Index

(5) Word Listening Test

(6) CES – D Scale

6.556*** (1.761)

17.271*** (4.402)

0.533** (0.241)

1.538** (0.746)

1.061** (0.532)

0.506 (0.485)

0.756*** (0.142)

1.639*** (0.321)

0.040** (0.019)

0.109* (0.060)

0.061 (0.043)

0.017 (0.039)

0.380** (0.149) 0.222 (0.187) 0.002 (0.001) 0.769*** (0.154) 0.455*** (0.115) 0.040 (0.186) 0.264** (0.105) 1.028*** (0.267)

9.607*** (0.370) 0.705 (0.429) 0.005* (0.003) 2.206*** (0.384) 0.389 (0.271) 0.034 (0.452) 0.539** (0.242) 1.330** (0.641)

0.014 (0.020) 0.046* (0.026) 0.000** (0.000) 0.056*** (0.021) 0.060*** (0.016) 0.126*** (0.026) 0.055*** (0.014) 0.109*** (0.037)

0.611*** (0.063) 0.127 (0.079) 0.001** (0.001) 0.365*** (0.065) 0.106** (0.049) 0.254*** (0.079) 0.309*** (0.045) 0.566*** (0.113)

0.544*** (0.045) 0.155*** (0.057) 0.002*** (0.000) 0.462*** (0.047) 0.020 (0.035) 0.033 (0.056) 0.204*** (0.032) 0.216*** (0.081)

0.388*** (0.041) 0.067 (0.052) 0.001 (0.000) 0.122*** (0.043) 0.040 (0.032) 0.369*** (0.051) 0.177*** (0.029) 0.260*** (0.074)

2.139*** (0.293) 0.352*** (0.076) 0.023*** (0.006) 0.071*** (0.010) 0.441*** (0.167) 0.037 (6.561) 4970

5.633*** (0.743) 0.847*** (0.177) 0.054*** (0.014) 0.132*** (0.022) 1.822*** (0.353) 27.217* (15.156) 4970

0.103** (0.040) 0.028*** (0.010) 0.001 (0.001) 0.002 (0.001) 0.045* (0.023) 4.790*** (0.900) 4970

0.454*** (0.124) 0.122*** (0.032) 0.003 (0.002) 0.005 (0.004) 0.130* (0.071) 6.957** (2.780) 4970

0.221** (0.089) 0.068*** (0.023) 0.005*** (0.002) 0.008*** (0.003) 0.262*** (0.050) 2.542 (1.982) 4970

0.188** (0.081) 0.038* (0.021) 0.001 (0.002) 0.001 (0.003) 0.108** (0.046) 4.161** (1.809) 4970

Note: The table presents fuzzy regression discontinuity (RD) estimates of the effect of self-reported type 2 diabetes on physical and mental health. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

A. Gaggero / Economics and Human Biology 37 (2020) 100830

13

Table A7 Instrumental Variable Estimates of Physical and Mental Health. Physical Health:

Reduced BMI Running Variable: FBG (mmol/L) Female [0,1] Years of Age Years of Age2 Higher Education [0,1] Family Size Married [0,1] Log-equivalised HH Income Smoking [0,1] Biomarkers: Haemoglobin A1c (%) LDL Cholesterol (mmol/L) Systolic Pressure (mm Hg) Diastolic Pressure (mm Hg) Taking Medication for: T2D [0,1] High Cholesterol [0,1] Hypertension [0,1] Time dummy Constant Observations

Mental Health:

(1) ADL Index

(2) Body Mobility Index

(3) Word Listening Test

(4) CES – D Scale

0.076** (0.033)

0.250** (0.104)

0.146 (0.093)

0.091 (0.075)

0.017 (0.018) 0.014 (0.027) 0.062** (0.031) 0.001** (0.000) 0.002 (0.032) 0.026 (0.032) 0.130*** (0.036) 0.035** (0.015) 0.187*** (0.057)

0.078 (0.063) 0.513*** (0.083) 0.190** (0.090) 0.002*** (0.001) 0.169* (0.097) 0.007 (0.086) 0.265*** (0.097) 0.242*** (0.051) 0.822*** (0.176)

0.044 (0.053) 0.598*** (0.064) 0.186*** (0.067) 0.002*** (0.001) 0.354*** (0.083) 0.084 (0.062) 0.026 (0.067) 0.166*** (0.042) 0.364*** (0.131)

0.049 (0.041) 0.351*** (0.054) 0.094 (0.059) 0.001* (0.000) 0.049 (0.068) 0.001 (0.053) 0.374*** (0.061) 0.153*** (0.036) 0.352*** (0.126)

0.058 (0.042) 0.002 (0.013) 0.001 (0.001) 0.007** (0.003)

0.101 (0.136) 0.013 (0.038) 0.003 (0.003) 0.022** (0.009)

0.081 (0.112) 0.022 (0.029) 0.002 (0.003) 0.018** (0.008)

0.022 (0.090) 0.013 (0.025) 0.002 (0.002) 0.005 (0.006)

0.088 (0.120) 0.018 (0.047) 0.028 (0.054) 0.010 (0.026) 4.825*** (1.058) 4970

0.241 (0.389) 0.185 (0.133) 0.046 (0.164) 0.044 (0.075) 6.656** (3.089) 4970

0.336 (0.300) 0.095 (0.102) 0.177 (0.139) 0.193*** (0.062) 2.720 (2.249) 4970

0.188 (0.305) 0.187** (0.085) 0.009 (0.116) 0.089* (0.052) 3.908* (2.010) 4970

Note: The table presents instrumental variable estimates of the effect of type 2 diabetes diagnosis on physical and mental health, through a reduction in the body mass index. Robust standard errors in parentheses are clustered at the individual level. * p < 0.1, ** p < 0.05, *** p < 0.01.Source: English Longitudinal Study of Ageing (ELSA).

Appendix B. Supplementary data Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.ehb.2019.100830.

References Balia, S., Jones, A.M., 2008. Mortality, lifestyle and socio-economic status. J. Health Econ. 27 (1), 1–26. Barreca, A.I., Lindo, J.M., Waddell, G.R., 2016. Heaping-induced bias in regressiondiscontinuity designs. Econ. Inq. 54 (1), 268–293. Bhaskaran, K., Douglas, I., Forbes, H., Dos-Santos-Silva, I., Leon, D.A., Smeeth, L., 2014. Body-mass index and risk of 22 specific cancers: a population-based cohort study of 524 million UK adults. Lancet 384 (9945), 755–765. Cawley, J., 2015. The impact of obesity on wages. J. Hum. Resour. 30 (2), 451–474. Cerulli, G., Dong, Y., Lewbel, A., Poulsen, A., 2017. Testing stability of regression discontinuity models. Adv. Econometr. 38, 317–339 Emerald Publishing Limited. Contoyannis, P., Jones, A.M., 2004. Socio-economic status, health and lifestyle. J. Health Econ. 23 (5), 965–995. Demakakos, P., McMunn, A., Steptoe, A., 2010. Wellbeing in older ages: a multidimensional perspective. chapter 4). Financial Circumstances, Health and Well-being of the Older Population in England. The 2008 English Longitudinal Study of Ageing.

Dong, Y., Lewbel, A., 2015. Identifying the effect of changing the policy threshold in regression discontinuity models. Rev. Econ. Stat. 97 (5), 1081–1092. Gaggero, A., 2019. The Effect of High Cholesterol Diagnosis on Lifestyle Behaviours: A Matching Estimator Approach. Working Paper. . Glechner, A., Keuchel, L., Affengruber, L., Titscher, V., Sommer, I., Matyas, N., Wagner, G., Kien, C., Klerings, I., Gartlehner, G., 2018. Effects of lifestyle changes on adults with prediabetes: a systematic review and meta-analysis. Prim. Care Diabetes 12 (5), 393–408. Hahn, J., Todd, P., Van Der Klaauw, W., 2001. Identification and estimation of treatment effects with a regression-discontinuity design. Econometrica 69 (1), 201–209. Han, E., Norton, E.C., Stearns, S.C., 2008. Weight and wages: fat versus lean paychecks. Health Econ. 18, 535–548. Huang, X., Pan, J., Chen, D., Chen, J., Chen, F., Hu, T., 2016. Efficacy of lifestyle interventions in patients with type 2 diabetes: a systematic review and metaanalysis. Eur. J. Intern. Med. 27, 37–47. Jahangir, E., De Schutter, A., Lavie, C.J., 2014. Low weight and overweightness in older adults: risk and clinical management. Prog. Cardiovasc. Dis. 57 (2), 127–133. Katz, S., Ford, A., Moskowits, R., Al, E., 1963. Studies of illness in the aged the index of ADL: a standardized measure of biological and psychosocial function. JAMA 185 (12), 914–919. Kenkel, D., 1995. Should you eat breakfast? Estimates from health production functions. Health Econ. 4 (1), 15–29. Kim, H.B., Lee, S.A., Lim, W., 2019. Knowing is not half the battle: impacts of information from the National Health Screening Program in Korea. J. Health Econ. 65, 1–14.

14

A. Gaggero / Economics and Human Biology 37 (2020) 100830

Kovesdy, C.P., Furth, S., Zoccali, C., World Kidney Day Steering Committee, 2017. Obesity and kidney disease: Hidden consequences of the epidemic. Indian J. Nephrol. 27 (2), 85–92. Kuk, J.L., Katzmarzyk, P.T., Nichaman, M.Z., Church, T.S., Blair, S.N., Ross, R., 2006. Visceral fat is an independent predictor of all-cause mortality in men. Obesity 14, 336–341. Lim, E.L., Hollingsworth, K.G., Aribisala, B.S., Cheng, M.J., Mathers, J.C., Taylor, R., 2001. Reversal of type 2 diabetes: normalisation of beta cell function in association with decreased pancreas and liver triacylglycerol. Diabetologia 54 (10), 2506–2514. Lindeboom, M., Lundborg, P., Van Der Klaauw, B., 2010. Assessing the impact of obesity on labor market outcomes. Econ. Hum. Biol. 8 (3), 309–319. Lyness, J., Noel, T., Cox, C., King, D., Conwell, Y., Caine, E., 1997. Screening for depression in elderly primary care patients. A comparison of the Center for Epidemiologic Studies-Depression Scale and the Geriatric Depression Scale. Arch. Intern. Med. 157 (4), 449–454. Mazzonna, F., Peracchi, F., 2012. Ageing, cognitive abilities and retirement. Eur. Econ. Rev. 56 (4), 691–710. McCrary, J., 2008. Manipulation of the running variable in the regression discontinuity design: a density test. J. Econom. 142 (2), 698–714. Milanovi
c, Z., Panteli
c, S., Trajkovi
c, N., Sporiš, G., Kosti
c, R., James, N., 2013. Agerelated decrease in physical activity and functional fitness among elderly men and women. Clin. Interv. Aging 8, 549–556. Morris, S., 2007. The impact of obesity on employment. Labour Econ. 14 (3), 413–433.

Moscoe, E., Bor, J., Bärnighausen, T., 2015. Regression discontinuity designs are underutilized in medicine, epidemiology, and public health: a review of current and best practice. J. Clin. Epidemiol. 68 (2), 122–133. Radloff, L.S., 1977. The CES-D Scale: a self-report depression scale for research in the general population. Appl. Psychol. Meas. 1, 385–401. Seuring, T., Goryakin, Y., Suhrcke, M., 2015. The impact of diabetes on employment in Mexico. Econ. Hum. Biol. 18, 85–100. Seuring, T., Serneels, P., Suhrcke, M., 2019. The impact of diabetes on labour market outcomes in Mexico: a panel data and biomarker analysis. Soc. Sci. Med. 233, 252–261. Slade, N.A., 2012. Health investment decisions in response to diabetes information in older Americans. J. Health Econ. 31, 502–520. https://www.bgs.org.uk/resources/diabetes. Thistlethwaite, D.L., Campbell, D.T., 1960. Regression-discontinuity analysis: an alternative to the ex post facto experiment. J. Educ. Psychol. 51 (6), 309–317. World Health Organization 2016. Global report on Diabetes. Zaninotto, P., Falaschetti, E., 2011. Comparison of methods for modelling a count outcome with excess zeros: application to Activities of Daily living (ADL-s). J. Epidemiol. Community Health 65 (3), 205–210. Zhao, M., Konishi, Y., Glewwe, P., 2013. Does information on health status lead to a healthier lifestyle? Evidence from China on the effect of hypertension diagnosis on food consumption. J. Health Econ. 32 (2), 367–385. Zivin, K., Llewell, D., Lang, I., Vijan, S., Kabeto, M., Miller, E., Langa, K., 2010. Depression among older adults in the United States and England. Am. J. Geriatr. Psychiatry 18 (11), 1036–1044.